To forecast time series future we have a lot of approaches. One them is to utilize neural networks (Deep Learning) like: Long-Short Term Memory (LSTM) or gated Recurrent Unit (GRU). The second one is to use non-machine learning approach including: ARMA, ARIMA, SARIMA, GARCH and so on to predict time series future steps.
Based on my experience LSTM and GRU are more accurate than traditional methods like ARMA family methods.
You can seeking for time series perdition using LSTM or GRU in GitHub or YouTube to fine codes and videos to learn it.
You can do that simply. You can try multiple ways but assume you know common habit of customer, like how much he spend per month on average. You put it and from curve you will be able to predict how much each month he will spend for next few years. Sum all that in disired period of time and you will know customer value over that time. You can use tons of methods and most of them will work (depending if data has trend or not + seasonal components). I would say Neural networks, Arima and similar models might be best for this tast.